ınbul 2004 
  
ight 
(level 1A) 
082 
-09-24 
54:39 
0065 
 x12000 
171756 
3 
t 
set of the 
projection 
six ground 
while the 
similarly, a 
the parallel 
while the 
square root 
adjustment 
in Tables 3 
resents the 
rdinates of 
model, as 
n Tables 3 
he first and 
at the first 
d with the 
lity of the 
acquisition 
parameters, 
en used to 
- geometry. 
absence of 
value of the 
heck points 
es 3 and 4, 
verifies the 
ting the y- 
ph covering 
in Figure 3. 
upports the 
normalized 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004 
Finally, the normalized scene coordinates as well as the derived 
parallel projection parameters for the normalized scenes are 
used in a space intersection procedure to derive the ground 
coordinates of corresponding object points. The derived ground 
coordinates for the check and control points, associated with the 
first and second datasets, are compared to the true values and 
reported in Tables 3 and 4, respectively. Similar to the analysis 
of the quality of fit, the first dataset is providing more accurate 
results (roughly half a pixel) when compared with the second 
one. The poor planimetric quality of the derived coordinates 
from the second dataset confirms the earlier prediction 
regarding a problem in the provided SPOT-5 data. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Parameter Value 
c, Left scene (pixels) 0.3 
c, Right scene (pixels) 0.2 
GCP Mean |P,| (pixels) 0.0 
in Mean,y * Std, (m) 0.000 + 2.340 
| Mean; + Std, (m) 0.000 + 0.858 
Check | Mean |P,| (pixels) 1.0 
points | Meanyy + Stdyy (m) 3.009 + 6.520 
errors | Meanz + Std, (m) 2.895 + 7.453 
Table 3: Results for the first dataset 
Parameter Value 
c, Left scene (pixels) 3:7 
c, Right scene pixels) 39 
GCP Mean |P (pixels) : 0.0 
S Meanyy € Std, (m) 0.000 + 18.605 
Meanz + Std, (m) 0.000 + 2.385 
Check | Mean |P,| (pixels) 0.0 
points | Meanyy + Stdyy (m) 31.436 + 64.985 
errors | Meany + Std; (im) 4.610 + 7.623 
  
  
  
  
Table 4: Results for the second dataset 
Since this study is mainly focusing on the evaluation of the 
quality of derived DEMs from SPOT-5 scenes, the remainder of 
the experimental results section is dedicated to the scenes 
captured over Belgium, dataset 2. A sub-region from the 
normalized scenes has been cropped as shown in Figure 4. The 
DEM generation process starts with the application of Fórstner 
interest operator to extract distinct points from the sub-scenes. 
A sample of the extracted points can be seen in Figures 5 and 6. 
The extracted points are then used as candidates for the 
matching procedure according to the proposed strategy in 
section 3. The results of the matching process have been 
visually inspected and we found out that only five points out of 
1876 were wrongly matched (a success rate of 99.7%). Figures 
5 and 6 show samples of correctly and wrongly matched points, 
respectively. 
Finally, the derived ground coordinates are used in an 
interpolation procedure to produce a high density DEM over the 
area shown in Figure 4. A shaded relief view of such DEM is 
displayed in Figure 7. 
Figure 6. Examples of correctly matched points as represented 
    
  
      
    
    
a) Left scene 
   
hn d 
Right scene 
b) 
Figure 3: Normalized scenes 
(12146x6416) for dataset 2 Figure 4: Cropped 
normalized scenes 
* 5 + 
E dud an dE. + = a ven LÀ = 
Figure 5. Examples of mismatched points as represented by the 
centres of boxes in the left and right scenes 
* Es 
+ wd " = # 
+ E: E + E 3 
by the centres of boxes in the left and right scenes